Adjusted Random Effect Block Bootstraps for Highly Unbalanced Clustered Data

Abstract

Clustered data arise naturally in many scientific and applied research settings where units are grouped within clusters. They are commonly analyzed using linear mixed models to account for within-cluster correlations. This article focuses on the scenario in which cluster sizes might be highly unbalanced and proposes a proportional random effect block bootstrap and a modified random effect block bootstrap, which are applicable in such cases and accommodate general distributions of random effects and error terms. These methods generalize the random effect block bootstrap, originally designed for the balanced case, and can be used for inference on parameters of linear mixed models or functions thereof. Both proposed bootstraps are shown to enjoy Fisher consistency under general cluster sizes, while the original random effect block bootstrap is consistent only for balanced clusters. Simulations demonstrate strong finite sample inferential performance of the proposed bootstraps relative to the random effect block bootstrap and other existing bootstrap methods for clustered data. Application to the Oman rainfall enhancement trial dataset, with cluster sizes ranging from 1 to 58, shows improved bootstrap confidence intervals using the proposed bootstraps over the random effect block bootstrap and a statistically significant effect of the ionization technology on rainfall.